Imprimir Republish


Checking citations with artificial intelligence

Scientists from the University of Illinois Urbana-Champaign, USA, received a US$300,000 grant to develop tools capable of assessing whether citations of scientific articles are reliable. The citations in a scientific paper’s bibliography make reference to previously published articles as a means of providing context and theoretical support. “A recent meta-analysis showed that 25.4% of medical articles contained a citation error,” said computer scientist Halil Kilicoglu, one of the researchers involved in the project, according to the University of Illinois website.

Kilicoglu gave an example of a single-paragraph correspondence published in the New England Journal of Medicine. “Inaccurate citations of a letter published in 1980 may have contributed to the opioid crisis,” he said. The letter claimed that cases of dependence among individuals with no history of addiction after taking narcotic drugs while in hospital in the US were very rare: at just four cases documented from a total of 11,800 patients. This correspondence was cited 608 times, of which 80% did not mention that the patients were in hospital under close monitoring, while others distorted the conclusions. The claim that addiction was unlikely encouraged the prescription of opioids—painkillers to which millions of Americans are now addicted and that were responsible for 183,000 deaths between 1999 and 2015.

Kilicoglu’s group plans to develop natural language processing and artificial intelligence models to help authors, reviewers, and journal editors verify the accuracy and completeness of citations in biomedical papers.